Acquiring a Machine Learning Data Set for Structural Health Monitoring of Hydrogen Pressure Vessels at Operating Conditions using Guided Ultrasonic Waves
Houssam El Moutaouakil,,Christian Fuchs,Enes Savli,Jan Heimann,Jens Prager,Jochen Moll,Kilian Tschöke,Octavio A. Márquez Reyes,Oliver Schackmann,Vittorio Memmolo,Tizian Schneider,,,,,,,,,,
DOI: https://doi.org/10.58286/29754
2024-07-01
Abstract:Hydrogen is an energy source of increasing importance. As hydrogen is very reactive to air and needs to be stored under high pressure, it is crucial to provide safe transportation and storage. Therefore, structural health monitoring, based on guided ultrasonic waves and machine learning methods, is used for Composite Overwrapped Pressure Vessels (COPVs) containing hydrogen. To acquire data that allows robust detection of COPV defects, there are two main process parameters to consider. These are the pressurization of the vessel and the temperature conditions at the vessel. This paper will focus on the derivation of a design of experiment (DoE) from the needs of various validation scenarios (e.g. concerning pressure, temperature or excitation frequency). Practical limitations must be considered as well. We designed experiments with multiple reversible damages at different positions. A network of 25 transducers, structured as five rings with five sensors in one line, is installed on a vessel. Guided ultrasonic waves are used via the pitch-catch procedure, which means that the transducers act pairwise as transmitter and receiver in order to measure all transmitterreceiver combinations. This leads to 600 signal paths, recorded by a Verasonics Vantage 64 LF data acquisition system. Finally, the influences of temperature and pressure within the acquired data set are going to be visualized.